Struct sparrow::scanner::booster::Boosting

pub struct Boosting { /* fields omitted */ }

The boosting algorithm. It contains two functions, one for starting the network communication, the other for starting the training procedure.

Methods

impl Boosting

pub fn new(
    exp_name: String,
    init_model: Tree,
    num_trees: usize,
    num_splits: usize,
    num_features: usize,
    min_gamma: f32,
    training_loader: BufferLoader,
    bins: Vec<Bins>,
    max_sample_size: usize,
    default_gamma: f32,
    save_process: bool,
    save_interval: usize
) -> Boosting

Create a boosting training class.

• num_trees: the number of boosting iteration. If it equals to 0, then the algorithm runs indefinitely.
• training_loader: the double-buffered data loader that provides examples to the algorithm. over multiple workers, it might be a subset of the full feature set.
• max_sample_size: the number of examples to scan for determining the percentiles for the features.
• default_gamma: the initial value of the edge gamma of the candidate valid weak rules.

pub fn enable_network(&mut self, name: String, port: u16)

Enable network communication. name is the name of this worker, which can be arbitrary and is only used for debugging purpose. port is the port number that used for network communication.

pub fn training(&mut self, prep_time: f32)

Start training the boosting algorithm.